Research On Robust Visual Tracking Algorithm Under Complex Environments

Visual object tracking has always been an important and challenging research topic in the field of computer vision,it provides support for further behavioral anal-ysis by locating and tracking specific target in the video.Object tracking involves many fields,including machine learning,image processing,pattern recognition,com-puter technology and so on.In recent years,with the rapid development of computer technology and modern sensor technology,object tracking technology have a good prospect in lots of industries,such as the video surveillance,intelligent transportation,human-computer interaction,behavior analysis,medical diagnosis,military guidance and so on,and more and more researchers are engaged in this area.Although there has been significant progress for object tracking in the past decades,however,developing a robust,real-time and accurate tracking algorithm to meet actual task is still a challeng-ing problem due to many challenging factors(eg.illumination change,pose change,scale,occlusion,rotation,motion blur,background clutter).In this paper,we focus on the problem of single-target tracking based on detec-tion.This paper first introduces the research background,the development history of the domain and several classical mainstream algorithms.After a lot of analysis and comparison,two robust algorithms are proposed.Through a lot of experimental analysis and demonstration,and with a number of comparative cutting-edge tracking algorithm from a number of aspects of comparative analysis,which highlights the ad-vantages of the proposed algorithm.The main contents and innovations of this paper are summarized as follows:1?In this paper,the Locality-constrained Linear Coding(LLC)algorithm is in-corporated into the object tracking framework.Firstly,we extract local patches within a candidate and then utilize the LLC algorithm to encode these patches.Based on these codes,we exploit pyramid max pooling strategy to generate a richer feature histogram.The feature histogram which integrates holistic and part-based features can be more discriminative and representative.Besides,an occlusion handling strategy is utilized to make our tracker more robust.Finally,an efficient graph-based manifold ranking al-gorithm is exploited to capture the relevance between target templates and candidates.For tracking,target templates are taken as labeled nodes while target candidates are taken as unlabeled nodes,and the goal of tracking is to search for the candidate that is the most relevant to existing labeled nodes by manifold ranking algorithm.Exper-iments on challenging video sequences have demonstrated the superior accuracy and robustness of the proposed method in comparison to other state-of-the-art baselines.2?An boosting learning algorithm based on the correlation filter is proposed.Over these years,Correlation Filter-based Trackers(CFT-s)have aroused increasing interests in the field of visual object tracking,and have achieved extremely compelling results in different competitions and benchmarks.Nevertheless,they only employ one feature and a single kernel,so they are usually not robust in complex scenes.In this paper,we derive a multi-feature and multi-kernel correlation filter based tracker which fully takes advantage of the invariance-discriminative power spectrums of vari-ous features and kernels to further improve the performance.A novel bootstrap learning method is utilized to obtain a strong classifier by fusing these weak kernel correlation filters(KCFs).Moreover,a new target scale estimation strategy is incorporated into our framework.The efficient and effective scale estimation method is based on target dictionary representation.The proposed method is tested on several videos and com-pared with seven state-of-the-art methods.Experimental results have provided further support to the effectiveness and robustness of the proposed method.